Is it better than DALL-E 2? | How does Imagen Actually Work?

After GLIDE and DALL-E 2, we have a new image generation model: Imagen! Like its predecessors, Imagen also uses diffusion models to achieve great results. In this video, let’s learn why Imagen is special, what its architecture looks like and how it creates the photorealistic images it does. Here is the article by Ryan O’Connor on Imagen: Our video on Diffusion Models: Article on Diffusion Models: 00:00 Introduction 00:43 Why is Imagen special? 01:17 The Architecture 01:47 Text Encoder 03:17 Image Generator 05:27 Classifier-free Guidance 07:04 Super Resolution Models 07:37 Model Evaluation 08:34 Wrap-up Is Imagen better than DALL-E 2? It is hard to answer since both Imagen and DALL-E 2 are not publicly available but from the published results, it looks like both of these models perform at a very similar level. They each have their own pros and cons, of course. How does Imagen work? Imagen is mainly based on a language model for caption understanding and a diffusion model for image generation. Is Imagen open source? Not yet. Google has decided not to release Imagen for public use before there are more safeguards in place. ▬▬▬▬▬▬▬▬▬▬▬▬ CONNECT ▬▬▬▬▬▬▬▬▬▬▬▬ 🖥️ Website: 🐦 Twitter: 🦾 Discord: ▶️ Subscribe: 🔥 We’re hiring! Check our open roles: ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ #MachineLearning #DeepLearning